Research

Access the list of my publications

I am an associate researcher at IRIT, within the SIG team, under the supervision of Olivier Teste.

Current research themes

My current research mainly focuses on ontology merging and knowledge alignment, in particular within the framework of the international OAEI (Ontology Alignment Evaluation Initiative) campaign.

Beyond algorithmic and formal aspects, I am especially interested in issues related to explainability and understanding the results produced by automated systems:

• how to explain a correspondence or alignment produced by an algorithm;
• how to verify or justify a model resulting from machine learning;
• how to make these results interpretable and usable by humans, especially in sensitive or critical contexts.

These topics, which lie at the intersection of data mining and knowledge engineering, are nowadays often associated (in a deliberately simplified way) with what is commonly referred to as artificial intelligence (AI).
However, my interest lies less in the buzzword than in the understanding, control, and responsibility of the systems that emerge from it.

Previous work

PhD research

During my PhD, I worked on issues related to disability, particularly visual impairment and text input on mobile devices.
My research aimed at better understanding the real difficulties faced by users, and at proposing more accessible interaction methods, taking cognitive, motor, and perceptual constraints into account.

This work followed a user-centered approach, combining experimentation, prototyping, and evaluation, with particular attention paid to the robustness and simplicity of interactions.
This research was conducted under the supervision of Mathieu Raynal.

Sentiment analysis (emotion mining)

Before that, my first research experience focused on sentiment analysis (emotion mining), during an internship at the LIA laboratory of EPFL.

This work addressed the analysis of online social interactions, combining network analysis and textual content analysis.
In particular, it highlighted the limitations of purely topological approaches (centrality, degree, etc.) for characterizing user behavior, and demonstrated the value of incorporating information derived from the emotional content of messages.

Special attention was given to the detection of antagonistic behaviors and their impact on community dynamics, opening perspectives toward behavior analysis, automatic moderation, and the study of complex social interactions.

Contact

For any question or collaboration related to research, you can contact me at the following address:

📧 Philippe.Roussille@irit.fr